emmacui

Cofounder and founding managing partner@longhashvc. DM open for founders. Personal opinion only, non financial advice. Sloan fellow @StanfordGSB. ex McKinsey

[emmacui @emmacui]

@Yuchenj_UW The cost of attack is dropping faster than most startups' security budgets are growing. Feels like a real opening for AI-native security built for small teams, not just enterprises

[emmacui @emmacui]

@ycombinator @MaxJunestrand @siggelabor @WeAreLegora Legal is a great test of vertical AI: high stakes, lots of documents, and buyers who care about trust more than demos. Getting 140k professionals to do their real work in the product is the moat. Congrats to the Legora team

[emmacui @emmacui]

7/ If you run finance or ops at a company with agents in production, how are you budgeting for them? I'd love to compare notes.

[emmacui @emmacui]

3/ On cost, their view is that token prices keep falling and usage rises to match, so total spend stays flat. Same story Reuters reported this week: OpenAI told investors price cuts on cheaper models boosted usage.

[emmacui @emmacui]

6/ So I think the next control layer looks less like a seat license and more like a spend policy: per-agent budgets, approval thresholds, audit trails. Startups like AIsa, which Tether just backed, are already selling exactly that.

[emmacui @emmacui]

2/ Their setup is simple: a monthly cap per employee for AI coding tools (about $200), with senior and key roles whitelisted with no cap. Many people just buy their own tools and never expense them.

[emmacui @emmacui]

1/ Everyone worries about runaway AI bills. A startup CFO I interviewed this weekend for my Stanford research on AI spend told me the budget is the easy part. Data is the hard part. 🧵

[emmacui @emmacui]

@a16z Pairing force feedback with vision feels like the missing piece for manipulation. Vision alone can't tell you how hard you're gripping. Curious whether the moat ends up being the hand hardware or the contact data it generates

[emmacui @emmacui]

@TimDraper From the investor side, the weeks right after the wire say a lot. The money should speed up a plan that already exists, not start the planning. Agree on the celebration point too: one night, then back to work.

[emmacui @emmacui]

5/ When generating output gets cheap, the value moves to deciding what's worth verifying. Where does verification break down in your AI workflow?

[emmacui @emmacui]

2/ Trust is what's holding AI back in critical systems. In her words: "They don't trust vibe coding, so that blocks AI deployment." Formal verification goes straight at that: the code ships with a proof that it does what it's supposed to.

[emmacui @emmacui]

4/ She closed with a big question about AI-generated math humans can't fully follow: "is math cancer biology or is math chess?" In other words, is it valuable for its own sake, or a tool for something else?

[emmacui @emmacui]

1/ Today I heard @CarinaLHong, founder of @axiommathai, speak at @DigEconLab on "AI That Can Prove It." Two ideas I keep thinking about 🧵

[emmacui @emmacui]

The intellectual horsepower in this photo is crazy!

[Alexandr Wang @alexandr_wang]

got the HRT gang back together!

[emmacui @emmacui]

@brian_armstrong Agents trading perps is where spend controls stop being a nice-to-have. Curious what guardrails come built in: position limits, leverage caps, a human approval step above a certain size?

[emmacui @emmacui]

Sources: Reuters, Axios and Breakingviews, all Oct 15.

Curious where people land: is 45% QoQ at this scale a slowdown, or just the new normal?

[emmacui @emmacui]

Anthropic just grew revenue about 45% in a single quarter, to $16.8B, and told investors it was profitable on an adjusted basis.

The reported reaction: maybe the IPO is worth about $1.5T, not $2T.

That's what it looks like to be priced on the curve, not the business.

[emmacui @emmacui]

For founders building on these models, that's mostly good news, because your input costs keep dropping.

The harder question is whether your own pricing holds when your competitors' costs drop too.